Sentence-Length Burstiness as a Cross-Disciplinary and Cross-Model Signal of AI Rewriting

Working paper and open per-text data. We compared 60,779 human-written academic texts with rewrites of the same texts generated by eight AI model configurations across six model families, so that every burstiness change is caused by the model rather than by topic or content. Burstiness is measured in its simplest form, the standard deviation of sentence lengths, with the coefficient of variation (CV) as the length-adjusted primary statistic. Human academic writing in this corpus has a mean CV of 0.449; the AI rewrites average 0.376, and 79.3 percent of all rewrites are flatter than their human written source. The flattening is universal across models but varies eight times in strength (DeepSeek -0.234, 98.5 percent flatter; Grok -0.032). A naive CV threshold that catches 62 percent of rewrites also falsely flags 39 percent of genuine human texts, concentrated on natural-science and engineering authors. Burstiness is a strong population-level signal but an unsafe individual-level verdict. Files: per-text sentence statistics for all 121,558 text sides (opaque pair identifiers; corpus, model, discipline, side, word and sentence counts, mean, standard deviation, and CV of sentence length), the example-pair sentence lengths, all analysis and figure code, and the paper PDF. The corpus pair texts are not included. Paper page: https://textpulse.ai/research/ai-burstiness-sentence-length